Some folks here have previously worked with our MegaDetector model for categorizing camera trap images as person/animal/vehicle/empty; we are excited to announce MegaDetector version 5.0, which provides a significant speedup (3x-4x) over MegaDetector v4, and - hopefully - increased accuracy in some of the scenarios that challenged MDv4.
MDv5 represents a change in architecture, but more importantly, an expansion of our training data to include some of the situations that make MDv4 sad, particularly human-made objects that overlapped with animals, like bait posts and traps. You can read a bit more about our training data here.
Looking forward to feedback from the community on where MDv5 shines, fails, or anywhere in between!
29 June 2022 11:07am
Great news Dan! We will be giving the model a go on some of our new eradication projects soon